Research map / Adversarial robustness and security / Attacks on text and graphs
ECGadv: Generating Adversarial Electrocardiogram to Misguide Arrhythmia Classification System
Huangxun Chen, Chenyu Huang, Qianyi Huang, Qian Zhang, Wei Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use smoothness instead of L2 to shape Type I perturbations (ablation) | ECGadv Type I · ECG target-class attacks | human preference for generated examples: 81.34% vs 18.66% | Improved |
| Add the smoothness metric to the Type I attack objective (ablation) | ResNet ECG classifier · Type I targeted attacks | targeted attack success rate: 97.22% -> 100%; 95.0% -> 100%; 99.55% -> 100% | Improved |
| Combine smoothness with an L2 penalty (ablation) | ECGadv Type I · Type I targeted attacks | targeted attack success rate: 100% | Improved |
| Use smoothness instead of Soft-DTW for perturbation similarity (ablation) | ECGadv Type I · ECG adversarial generation | CPU time per iteration: 12.28s -> 0.05s | Improved |
| Model ECG filtering and random time shifts during perturbation generation | ECGadv Type II · Type II targeted attacks across source-target classes | targeted attack success rate: 99.64% success rate | Improved |
| Reduce perturbation window size (ablation) | ECGadv Type II · Type II attacks, most source-target pairs | targeted attack success rate | Hurt |
| Use a shorter perturbation window for attacks targeting class O (ablation) | ECGadv Type II · A→O, N→O, and noise→O | targeted attack success rate | No clear effect |
| Apply the alternative Butterworth and notch filters instead of the generation-time rectangular filter (ablation) | ECGadv Type II · Type II targeted attacks | targeted attack success rate: 99.97%/99.96%; 100%/100%; 97.63%/97.45% | No clear effect |
| Use smoothness rather than smoothness plus L2 (ablation) | ECGadv Type I · ECG line-curve similarity | human preference: 59.67% vs 40.33% | Improved |
Closest papers by meaning
- Robustness and Visual Explanation for Black Box Image, Video, and ECG Signal Classification with Reinforcement Learning (AAAI 2024 · similarity 0.84)
- Boosting Masked ECG-Text Auto-Encoders as Discriminative Learners (ICML 2025 · similarity 0.81)
- H-Tuning: Toward Low-Cost and Efficient ECG-based Cardiovascular Disease Detection with Pre-Trained Models (ICML 2025 · similarity 0.79)
- ECG-Reasoning-Benchmark: A Benchmark for Evaluating Clinical Reasoning Capabilities in ECG Interpretation (NeurIPS 2026 · similarity 0.79)
- ECG-R1: Protocol-Guided and Modality-Agnostic MLLM for Reliable ECG Interpretation (ICML 2026 · similarity 0.79)
- Guiding Masked Representation Learning to Capture Spatio-Temporal Relationship of Electrocardiogram (ICLR 2024 · similarity 0.78)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
